Dr. Robert Rowe & Jonathan S. Abrams
Robert Auld opened the meeting by asking how many attendees were AES members. Almost all of them were. Mr. Auld noted that the next New York Section meeting, on December 8th, would be held at a new venue, the Hearst Building auditorium at 300 West 57th Street.
Jonathan Abrams then introduced Dr. Rowe, a composer with a special interest in music and cognition. Dr. Rowe has worked with the Institute of Sonology in Utrecht and with IRCAM in Paris, where he developed control level software for the 4X machine (a musically oriented computer). With this background he is especially qualified to investigate the use of computers for analyzing music and human perception of music, and much of his presentation centered around current developments in that field.
Dr. Rowe started by noting the limits of previous efforts to analyze musical content, based on examination of MIDI signals. The 20 year old MIDI standard, while still very useful, is limited in the amount of information it conveys about music—largely the timing, dynamics and pitch, but nothing about timbre for example. So now, efforts are being focused on analysis of recordings of actual music. The emphasis is on finding out how humans percieve music, and finding ways for computers to learn such comprehension.
A simple example is called specific loudness, which defines how loud something sounds to a person, rather than just the measured physical amplitude of the sound signal. A research company that Dr. Rowe works with, Source Tone, currently makes computer software plug-ins that use specific loudness parameters for measuring and modifying sound. Dr. Rowe demonstrated several of those plug-ins for the meeting.
A more complex problem is attempting to train computers to predict human emotional reactions to specific pieces of music. As part of this effort, Dr. Rowe and his associates have developed a Circumplex Model of Emotions which is used to collect and analyze data on human reactions to varied musical samples. The model can be plotted in two dimensions, showing valence (unpleasant to pleasant reactions) and arousal (calm to excited). The four quadrants possible with such a plot can encompass a wide range of listener reactions to a sample. Data collected from hundreds of subjects has then been used to develop empirical models of human perception of hundreds of musical samples. In turn, this data has been used to develop computer algorithms for predicting emotional reactions to specific pieces of music. According to Dr. Rowe, the accuracy of prediction of the software currently is better than 70 percent.
Finally, Dr. Rowe spoke briefly about using music as therapy, referring to a recent article in the Wall Street Journal about the use of music in treating Alzheimers patients.
Following Dr. Rowe's presentation there was a question and answer session, with many of the approximately fifty attendees participating.
